ELK Reasoner: Architecture and Evaluation

ELK is a specialized reasoner for the lightweight ontology language OWL EL. The practical utility of ELK is in its combination of high performance and comprehensive support for language features. At its core, ELK employs a consequence-based reasoning engine that can take advantage of multi-core and multi-processor systems. A modular architecture allows ELK to be used as a stand-alone application, Protégé plug-in, or programming library (either with or without the OWL API). This system description presents the current state of ELK and experimental results with some difficult OWL EL ontologies. 1 The System Overview The logic-based ontology language OWL is becoming increasingly popular in application areas, such as Biology and Medicine, which require dealing with a large number of technical terms. For example, medical ontology SNOMED CT provides formal description of over 300,000 medical terms covering various topics such as diseases, anatomy, and clinical procedures. Terminological reasoning, such as automatic classification of terms according to subclass (a.k.a. ‘is-a’) relations, plays one of the central roles in applications of biomedical ontologies. To effectively deal with large ontologies, several profiles of the W3C standard OWL 2 have been defined [11]. Among them, the OWL EL profile aims to provide tractable terminological reasoning. Specialized OWL EL reasoners, such as CEL [1], Snorocket [9], and jCEL [10], can offer a significant performance improvement over general-purpose OWL reasoners. This paper describes the ELK system.3 ELK is developed to provide high performance reasoning support for OWL EL ontologies. The main focus of the system is (i) extensive coverage of the OWL EL features, (ii) high performance of reasoning, and (iii) easy extensibility and use. In these regards, ELK can already offer advantages over other OWL EL reasoning systems mentioned above. For example, as of today, ELK is the only system that can utilize multiple processors/cores to speed up the reasoning process, which makes it possible to classify SNOMED CT in less than 10 seconds on a commodity hardware [4]. This paper presents an overview of the implementation techniques used in ELK to achieve high performance of classification, and provides an experimental evaluation of classification using ELK and related reasoners on some of the largest available OWL EL ontologies. 3 http://elk-reasoner.googlecode.com/ Command-line Client Reasoner OWL FSS Parser Protégé Plugin OWL API Bindings OWL